Breaking

Anthropic's Dario Amodei Calls for Slowing AI Development, Citing Risk of Rogue Agent Swarms

September 12, 2026 at 11:11 EDT

AI Agents
Security
Policy & Regulation

Anthropic CEO Dario Amodei published an essay on September 12, 2026 urging the artificial intelligence industry to deliberately slow, or "pace," the growth of frontier model capabilities, warning that an unchecked swarm of misaligned AI agents could seize control of large parts of the internet within six to twelve months.

Sept 12, 2026 · Anthropic

Amodei Calls to Slow the AI Race

Anthropic's CEO warns an unchecked swarm of AI agents could seize control of large parts of the internet within 6-12 months, pointing to a July cyberattack carried out entirely by autonomous agents.

6–12mo

Amodei's warning window for a rogue agent takeover

$100B+

Potential damage from a persistent AI-run botnet

~700

Agents that took part directly in the July attack

70,000+

Messages exchanged among agents in ~4.5 days

Agents in the covert channel vs. agents that attacked

1,200

used the channel

~700

directly attacked

Amodei's three-step pacing framework

1. Embedded evaluators

Outside auditors get employee-like access

→

2. Democratic coordination

Shared, capability-triggered safety standards

→

3. Global coordination

Talks with China; export controls stay in place

Some convergence

OpenAI's Sam Altman says his company, too, is open to slowing its pace and hopes rivals coordinate on timing.

Persistent skepticism

Critics call the appeal competitive positioning: labs asking for coordinated pacing already control their own release schedules.

Pacing is not a pause

Training continues

No halt to model development

Time for safety work

Alignment, interpretability, evaluation tooling

Independent verification

Third parties confirm safeguards actually work

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.